Implement custom fold for ZipSlice

[Owen-CH-Leung]

As per discussion in #755 , this PR implements custom fold logic for ZipSlices<T, U> and create benchmark test to measure the performance gain compared with the default fold.

Running the zipdot i32 zipslices fold benchmark using the default fold logic, the time was time: [42.749 ns 43.163 ns 43.782 ns]

Running the zipdot i32 zipslices fold benchmark using the custom fold logic, the time was time: [42.523 ns 42.621 ns 42.737 ns]

[Philippe-Cholet]

I was gonna say I was surprised by the unsafe use, especially for such time win but ZipSlices is not part of the itertools library, it's only part of benches.

[jswrenn]

I appreciate the help on #755. However, as @Philippe-Cholet suggests, our focus is on improving the performance of user-facing iterators. Specializing this particular fold adds maintenance cost without any performance gain (even internally).

[Owen-CH-Leung]

Ahh yesss I misread. I thought it's under src when I was revising it. Thanks both for your comments. I'll continue to see if custom fold can be implemented in other iterators